Earth Surface Processes and Landforms
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
SEDIMENT YIELD VARIABILITY IN THE UPPER YANGTZE, CHINA
XIXI LU1 AND DAVID L. HIGGITT2*,
Department of Geography, University of Western Ontario, London, Ontario, N6A 5C2, Canada
2
Department of Geography, University of Durham, Science Laboratories, South Road, Durham, DHI 3LE, UK
1
Received 30 November 1997; Revised 26 October 1998; Accepted 17 February 1999
ABSTRACT
The development and increasing availability of global environmental data sets provides an opportunity to examine
systematically the relationship between sediment yields and controlling catchment variables, employing Geographical
Information Systems. Few studies have attempted to harness such information to analyse variations in sediment yields
within large catchments. Sediment yields from 62 long-term gauging stations within the catchment of the Upper Yangtze
River, China, have been analysed in relation to variables describing hydrology, climate, topography and population density.
This analysis is particularly significant as the 106 km2 catchment area of the Upper Yangtze will shortly be impacted by the
world's largest dam scheme (the Three Gorges Project). There is a high degree of scatter in sediment yields because of
natural diversity in the catchment, but this scatter is reduced when the data are grouped according to tributary location,
catchment size and maximum elevation. Sediment yields generally increase with precipitation, runoff and population
density and decrease with elevation, but there is evidence of scale dependency and of variation between geographic regions
within the basin. The small number of variables used are capable of explaining the majority of variance in the comparatively
`natural' western tributaries but are less adequate in areas affected by large-scale agricultural activity. In future,
improvements in the resolution and accessibility of environmental data sets will allow more detailed analysis of regional
variability in sediment yield. Copyright # 1999 John Wiley & Sons, Ltd.
KEY WORDS: sediment yield; global environmental data sets; Upper Yangtze River; Three Gorges
INTRODUCTION
The Upper Yangtze, upstream of Yichang, Hubei Province, China, includes a diverse range of environments.
The Yangtze rises on the arid Qinghai-Tibet (Xizang) Plateau, where a large proportion of the land is at an
elevation above 4000 m, before descending into the fertile Sichuan Basin. The Upper Jinsha, Yalong, Dadu
and Min drain the mountainous west, the Tuo, Fu, Jialing and Qu drain the agricultural lands of Sichuan in the
east of the catchment, and the Wu drains the uplands of Guiyang province (Figure 1). The proposal to
construct the world's largest hydro-power scheme on the Yangtze (the Three Gorges Project, TGP) has
focused attention on the implications of soil erosion and fluvial sediment transport in the Upper Yangtze
basin. The life span of the scheme could be threatened by extensive sedimentation although the dam
engineers are confident that river regulation procedures can reduce this hazard (Qian et al., 1993).
Widespread evidence that the extent and magnitude of soil erosion has increased dramatically during the last
40 years (Edmonds, 1992; Wen, 1993) is not clearly matched by trends in the sediment load measured at
Yichang, a long-term gauging station a short distance downstream of the TGP dam site (Figure 1). Two key
questions for those involved in catchment management concern identification of the main sources of
sediment and their conveyance to the main Yangtze channel. A number of studies have examined temporal
and spatial patterns of sediment transfer in the Upper Yangtze (Gu et al., 1987; Gu and Douglas, 1989; Qian
et al., 1993; Zhou and Xiang, 1994; Higgitt and Lu, 1996), but there has been little attempt systematically to
examine the relationship between sediment yield and its controlling variables.
* Correspondence to: D. L. Higgitt, Department of Geography, University of Durham, Science Laboratories, South Road, Durham,
DHI 3LE, UK. E-mail: D. L. [email protected]
CCC 0197-9337/99/121077±17 $17.50
Copyright # 1999 John Wiley & Sons, Ltd.
1078
X. LU AND D. L. HIGGITT
Figure 1. The Upper Yangtze Basin. Major areas of level ground on the Qinghai±Tibet Plateau and the Chengdu Plain (part of the Sichuan
basin) are indicated. The inset shows the regional setting and the names of provinces in the area
Several attempts have been made to explain global and regional patterns of sediment yield in terms of
either climate and vegetation (Langbein and Schumm, 1958; Douglas, 1967; Wilson, 1973; Jansen and
Painter, 1974; Jansson, 1988) or topography (Milliman and Syvitski, 1992; Summerfield and Hulton, 1994).
Within China, Xu (1994) has proposed that national variations in sedimentation are adequately described by
the Langbein±Schumm model, but inspection of sediment load data within the Upper Yangtze suggests
greater complexity, reflecting the combination of topographic, hydroclimatic, lithological, land-use and soil
erodibility controls. Attempts to assess the relative importance of these controls in explaining patterns in
sediment yield have been hampered by the difficulty of obtaining sufficiently detailed, spatially distributed
information. However, the advent of global environmental data sets containing description of the
hydroclimatic, biological and geomorphological characteristics of the Earth (Ludwig and Probst, 1996)
offer the potential for extracting catchment variables for integration with sediment yield data. The resolution
of global data sets has tended to restrict previous analyses to the global scale, where individual catchments are
represented by single sediment yield variables (Summerfield and Hulton, 1994; Ludwig and Probst, 1996).
The application of such an approach to a large basin (>106 km2) containing a hierarchy of nested gauging
stations appears to have received limited attention.
To address this limitation, the principal aim of the paper is to examine the controls on sediment yield within
the Upper Yangtze basin, using catchment data extracted from various global data sets. Recognizing that
many relationships are widely scattered, attempts are made to reduce scatter through screening analyses based
on geographic location (tributary), catchment size and elevation. Previous work has described spatial and
temporal variations in sediment load (Higgitt and Lu, 1996; Lu and Higgitt, 1998a) and approaches to
mapping regional sediment yields (Lu and Higgitt, 1998b).
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1079
SEDIMENT YIELD VARIABILITY
DATA SOURCES AND METHODS
Sediment yield and runoff data
Sediment yield (SY) and runoff (RO) data are derived from a network of hydrographic stations throughout
the Upper Yangtze. Some river gauging records extend back to the 1930s, but the majority commence in the
1950s. The original records for each station provide information on station co-ordinates (latitude and
longitude), catchment area, mean monthly and annual water discharge and sediment load, and the magnitude
and date of occurrence of the maximum daily discharge. Load data refer to measured suspended sediment and
exclude any bedload. The annual measured sediment loads were recorded in units of 106 or 104 t depending
on the station basin area or measurement year and it is apparent that the compilation of records into yearbooks
has introduced some errors in these units (Higgitt and Lu, 1996). Recalculation of annual load from monthly
load data has enabled several of these transcription errors to be corrected. The recorded catchment areas and
locations of many stations are inconsistent within the measurement series and might reflect correction of
previous area measurement or a slight change of station location. Error is also introduced through the use of
daily or weekly measurements, rather than continuous monitoring, which is likely to underestimate sediment
discharges during peak flows (Waythomas and Williams, 1988).
Sediment load data were extracted from 250 stations in the Upper Yangtze for the period 1956±1987. As
the catchment experiences marked year-on-year climatic variations, it was decided to restrict analysis of the
relationship between sediment yield and potential controlling variables to those stations with 25 or more years
of record (Table I). Data for 56 stations which met this criterion were supplemented by a further six stations
from the Wu tributary. The resulting distribution of the stations comprises 17 in the Jinsha±Yalong
catchment, 15 in the Dadu±Min (including Tuo), 20 in the Jialing, six in the Wu and four in the catchment
area of the Main channel. Summary information describing these stations is listed in Table II.
Sediment outputs from catchments of different sizes are normally expressed as specific sediment yields (t
kmÿ2 aÿ1). There are two approaches to calculating specific sediment yields in hierarchical sub-catchments.
First, they can be calculated by deducting the sediment load at the immediate upstream station from that at the
gauging station, and then dividing this value by the incremental catchment area (Lajczak and Jansson, 1993;
Ozturk, 1996). This method is problematic at stations where there is a net loss of sediment load downstream
(for example, through floodplain deposition), as this will produce a negative value, especially in downstream
reaches of large rivers. The alternative approach, which is used here, is to express specific sediment yield as
total load divided by total catchment area upstream of the station, although this means that the problem of
spatial averaging is accentuated in the downstream direction. The influence of scale on relationships between
sediment yield and catchment variables requires further consideration.
Table I. Gauging station length of record and catchment areas
Numbers
Percentage
Catchment area
(km2)
Numbers
1±4
5±9
10±14
15±19
20±24
25±32
63
30
22
29
50
56
252
120
88
116
200
224
<100
100±1000
1000±10 000
10 000±100 000
100 000±1 000 000
>1 000 000
1
60
119
48
21
1
Total
250
100
Total
250
Measurement years
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1080
X. LU AND D. L. HIGGITT
Table II. Mean speci®c sediment yields of the selected sub-catchments
No.
Tributaries
Stations
1
6
16
23
25
29
30
31
32
36
42
64
67
69
70
71
73
81
83
84
85
86
91
92
94
97
98
105
116
125
128
129
133
140
149
153
154
156
160
164
165
168
171
178
179
180
182
183
189
190
195
197
200
201
206
207
208
220
236
242
243
250
Jinsha-Yalong
Zimenda
Shigu
Huatan(Qiaojia)
Ninnan
Qianxinqiao
Meigu
Pingshang
Hengjiang
Zhutuo
Dianwei
Qinkoutang
Lounin
Xiaodeshi
Anninqiao
Sunshuiguan
Manshuiwan
Wantan
Zengjianguan
Shaba
Jiangsheba
Zagunao
Shuangping
Yanliuping
Xinxinchang
Pengshan
Qinshuixi
Gaochang
Dajin
Shaping
Shanhuangmiao
Denyenyan
Lijiawan
Yunninzeng
Liuanyang
Wudu
Bikou
Sanleiba
Tinzhikou
Qinquanxian
Wusheng
Beipei
Bixi
Qilitou
Dunlin
Minyuantan
Guodukou
Jinbian
Luoduxi
Fujiangqiao
Guanyinchang
Shehong
Xiaoheba
Yachihe
Wujiangdu
Shinan
Wulong
Gongtan
Duntou
Chuntang
Shizhu
Wanxian
Yichang
Dadu-Min
Tuo
Jialing
Wu
Main channel
Copyright # 1999 John Wiley & Sons, Ltd.
DA (km2)
Years
Mean
(t kmÿ2 aÿ1)
SD
Min.
Max.
137 704
232 651
450 696
3074
2549
1607
485 099
14 781
694 725
120
2109
108 083
118 294
937
1596
3817
11 100
4486
7231
14 279
2404
4629
363
396
30 661
3330
135 378
40 484
75 016
6590
14 484
23 283
2071
19 206
14 288
26 086
29 247
61 089
5011
79 714
156 142
2124
6382
6462
736
31 626
2740
38 071
11 903
1933
23 574
29 420
16 541
26 496
50 791
83 035
58 346
6917
866 559
898
974 881
1005501
28
28
30
25
27
25
31
25
26
29
29
28
27
27
26
32
26
27
31
26
26
30
26
26
30
28
32
27
21
28
31
29
25
30
25
27
29
28
25
31
31
28
27
29
27
30
27
31
25
26
29
30
21
23
21
27
21
26
30
25
20
31
68
91
366
1191
65
1152
505
919
459
262
798
175
249
657
1770
975
973
124
344
307
284
387
812
1257
337
487
363
107
420
853
617
537
458
1706
1194
632
563
1027
598
928
955
661
586
1216
1108
630
395
760
992
228
705
650
886
498
345
390
366
94
518
719
503
524
38
42
133
764
40
653
177
367
102
245
364
79
114
372
1523
659
587
72
211
198
170
216
794
1242
157
230
160
53
146
440
365
351
420
1327
860
453
359
644
562
562
469
475
393
726
447
344
263
442
921
172
527
566
409
305
186
148
149
49
127
1013
122
98
9
30
221
330
22
473
260
429
296
20
272
77
107
238
350
263
389
38
94
75
63
145
23
139
131
86
167
31
189
204
73
94
46
273
223
79
110
196
96
123
189
55
72
273
423
102
31
94
139
14
43
59
193
9
17
134
70
17
333
157
346
361
139
182
707
3193
177
3103
1034
1629
668
931
1629
333
544
2159
7237
3042
2707
295
1034
915
810
1145
3695
5315
721
1346
897
266
732
1954
1571
1532
1513
6638
4660
1609
1381
2681
2821
2542
2284
1605
1389
3394
2024
1517
1022
1757
4056
726
2584
3121
1727
1131
746
730
719
176
823
5377
819
725
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
SEDIMENT YIELD VARIABILITY
1081
Catchment boundary delineation
Identification of boundaries to each sub-catchment is an essential stage in extracting spatially distributed
variables using a geographical information system (GIS). Given the co-ordinates of the gauging station, a
high resolution digital elevation model (DEM) can be employed to derive catchment boundaries and, hence,
most catchment morphometric variables. The Asian 30 arcsecond DEM has the highest resolution elevation
data available in the public domain for this part of China. The primary source of the DEM was a
generalization of the Level 1 Digital Terrain Elevation Data (DTED) produced by the US Defense Mapping
Agency. Elevation data for the areas of Asia which are not covered by the DTED were developed for the
Asian 30 arcsecond DEM, using the 1:1 000 000 scale Digital Chart of the World. The original 3 arcsecond
DTED data for Asia have been resampled to produce the Asia 30 arcsecond DEM, which yields an accuracy
of approximately 1 km across the latitude range of interest. Vertical resolution is 30 m.
Using the available DEM, some catchment boundaries were delineated using Arc/Info GIS (ESRI, 1994).
However, it is difficult to generate catchment boundaries in relatively flat areas such as the lower Jialing
tributary in the Chengdu Plain part of Sichuan Basin. Some catchments were therefore digitized from
1:1 500 000 maps. There is good agreement between the delineated/digitized catchment areas and the
recorded catchment areas, although the resolution of the DEM and topographic map is low. Discrepancies
between the delineated/digitized and the recorded areas were generally small. For example, differences of <5
per cent were identified for 40 (645 per cent) of the 62 catchments and differences of 5±10 per cent were
found for 19 (306 per cent) of the catchments. Only three catchments had differences as large as 10 to 15 per
cent (47 per cent).
Extraction of catchment variables
The following variables were extracted for each of the 62 sub-catchments defined by the catchment
boundaries: mean elevation (ME; m), basin relief (BR; m), mean slope (MS; degrees), population density
(PD; persons kmÿ2) and precipitation (PP; mm).
Mean elevation is defined by the arithmetic mean of the elevations of all cells within the catchment
boundary, while basin relief is the difference between maximum and minimum cell values. Mean, maximum
and minimum cell values for each catchment were obtained from a statistical file after clipping the 30
arcsecond DEM using the delineated/digitized catchment boundary. A slope grid was generated using slope
commands in Arc/Info, from which mean slope, defined by the mean of the slopes of all cells, was extracted
for each catchment. The slope was calculated based on maximum elevation changes within nine neighbouring
cells (ESRI, 1994). Clearly, at the resolution involved, slope is a rather crude indicator of topographic
variation.
Population density data were derived from the Asian Population Database (APD). The development of the
APD was supported by the United Nations Environment Programme/Global Resource Information Database
(UNEP/GRID) and the Consultative Group for International Agricultural Research (CGIAR). The Asian
population database is part of ongoing effort to improve global, spatially referenced demographic data
holdings. Such databases are useful for a variety of applications including strategic level agricultural research
and applications in the analysis of the human dimensions of global change. This project has pooled available
data sets, many of which had been assembled for the global demography project. For China, the data were
products of the 1992 census. Data from APD were downloaded from the Internet and converted into gridded
data using Arc/Info.
Precipitation data were derived from the Global Ecosystems Database (GED) version 1 (NOAA-EPA
Global Ecosystems Project, 1992) which is an integrated database related to global environmental and
ecological change. The database contains many useful map distributions, such as cultivation intensity and
soils, with resolutions varying from 10 10 minutes to 1 1 degree. Unfortunately, the resolutions for most
variables are poor across China, making them unsuitable for this scale of regional study. Consequently, only
annual and monthly average precipitation data, at a resolution of 05 05 degree (Legates and Willmost,
1992) were considered suitable for the current analysis. Average annual precipitation was defined as the mean
for all cells within the catchment boundaries.
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1082
X. LU AND D. L. HIGGITT
Figure 2. Distribution and range of mean specific sediment yield (t kmÿ2 aÿ1) in the Upper Yangtze Basin
RESULTS
This section is divided into three parts. First, the spatial distribution of the extracted data is examined and
general geographical patterns are described. Second, the interrelationships between the catchment variables
are examined as it is likely that there will be some collinearity between variables. Third, variation of sediment
yield with catchment variables is examined through grouping data by tributary, catchment size and maximum
elevation to reduce scatter.
Sediment yield variability and catchment variables
Mean annual specific sediment yields (t kmÿ2 aÿ1) for all Upper Yangtze stations with five or more years
of record are plotted in Figure 2. It is well known that specific sediment yield declines with catchment area as
opportunities for sediment storage increase (Walling, 1983). In an earlier paper, Higgitt and Lu (1996)
represented sediment yields as standardized residuals from the specific sediment yield±catchment area
regression. Nevertheless, the raw data illustrate the spatial pattern of sediment yields and indicate that the
gauging stations are distributed unevenly, with most located in the populated east and few in the mountainous
west. The upper Jialing and the lower Jinsha have much higher sediment yields than the upper Jinsha and
Yalong tributaries. Using standardization procedures, Higgitt and Lu (1996) found that the catchments of the
Jialing, Fu and Qu rivers were the dominant sediment source area, but showed that the relative importance of
Jialing declined during the 1970s, whereas the Qu became an increasingly important source throughout the
period.
Information on topography, slope, precipitation and population densities are displayed in Figure 3. The
topographical map (Figure 3 a) is produced from the Asian 30 arcsecond DEM based on five classes, 0±200,
200±500, 500±1000, 1000±3000 and >3000 m. It emphasizes the large area of high elevation on the Qinghai±
Tibet plateau and the very deep and narrow valleys draining it. A large number of the sub-catchments in the
sediment yield data set contain land above 3000 m, in contrast to previous studies of global or regional
controls on sediment yield which are skewed towards lower elevations (Milliman and Syvitski, 1992; Probst
and Amiotte-Suchet, 1992; Summerfield and Hulton, 1994).
The slope map indicates that the steep areas (>10 ) are mainly distributed on the western margins of the
Sichuan Basin and along the incised valleys of the Jinsha, Yalong and Dadu tributaries. There are two
relatively flat areas, the Chengdu Plain and Qinghai±Tibet plateau (Figure 1), where the dominant mean slope
angle is less than 1 . However, it should be noted that many flat areas (recorded as 0 ) are generated by the
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
Copyright # 1999 John Wiley & Sons, Ltd.
(b)
(c)
(d)
Figure 3. Distribution maps of mean cell values for (a) elevation (m), (b) slope (degrees), (c) mean annual precipitation (mm) and (d) population density (persons kmÿ2)
1083
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
SEDIMENT YIELD VARIABILITY
(a)
1084
X. LU AND D. L. HIGGITT
GIS due to problems inherent to the resolution of the DEM. This problem must be taken into account when
the slope is applied in sediment yield modelling.
The precipitation map also emphasizes the dramatic contrast between the arid/semi-arid northern Qinghai±
Tibet Plateau and the eastern part of the basin, where precipitation exceeds 1000 mm. The Qinghai±Tibet
plateau has a significant influence on atmospheric circulation not only regionally but perhaps over the entire
northern hemisphere (Ruddiman et al., 1989). The plateau constrains penetration of the monsoon, resulting in
a complex pattern of precipitation within the Upper Yangtze catchment.
Population densities are generally inversely related to elevation, with values ranging from less than 10
people kmÿ2 in the Qinghai±Tibet Plateau to >500 persons kmÿ2 in the Sichuan Basin, which has one of the
highest population densities in China. The total population of Sichuan province exceeds 01 billion. The
Jialing, Tuo and Wu flow through the higher population density areas.
Interrelationships between catchment variables
The GIS environment enables values for both individual cells (pixels) and sub-catchments to be calculated
efficiently. Examination of the interrelationships between these catchment variables precedes analysis of
their relationship to sediment yield. As many of the distributions of catchment variables do not closely
approximate to normal distributions, and because the interaction between variables may not be linear,
parametric statistical techniques are inapplicable. Results from a non-parametric Spearman's rank correlation
matrix are presented in Table III. As expected, mean elevation is significantly correlated with many other
variables. Mean slopes increase with mean elevation, whereas mean population density and precipitation
decrease with mean elevation. However, the relationships are more complex at pixel level (Figure 4). The
headwaters of the Jinsha and Yalong Rivers on the Qinghai±Tibet plateau are areas of gentle relief where the
mean slope generally increases with elevation up to 2500 m but then decreases. Population density decreases
sharply with elevation above 1000 m elevation. Elevation and precipitation are inversely related, but with a
high degree of scatter in lower elevation areas.
Modelling sediment yield variability
The high degree of spatial variability in sediment yields and catchment characteristics causes difficulty
when attempting to model controlling relationships using the whole data set. Considering the 62 selected
catchments, specific sediment yields are significantly correlated with only mean elevation and runoff (Table
III). The role of elevation as a direct control on sediment transport is questionable. Elevation, as a measure of
potential energy, has an influence on erosion potential but in this case appears to act as a surrogate for aridity
and human impact, both of which have negative correlations with elevation. The interdependence of variables
likely to influence sediment production and transport is problematic. Simple relationships between sediment
yield and a catchment variable, such as elevation, can reflect underlying influences. The use of multiple
regression procedures to establish the optimum explanation of sediment yield from a selection of catchment
variables must take into account the redundancy of each additional variable. In this case, stepwise procedures
were used to examine the improvement in explanation provided by each additional variable. Attempts to
explain controls on sediment yields are also subject to high degrees of scatter. Previous studies of sediment
yields have attempted to reduce this scatter by grouping data into suitable categories. In global-scale studies,
Milliman and Syvitski (1992) grouped sediment yield data by maximum elevation, while Ludwig and Probst
(1996) and Jansson (1988) used climate type. The present study groups the 62 basins on the basis of tributary,
basin size and maximum elevation.
Basin grouping based on tributary
Provision of an increased understanding of the controls on sediment production and output within each of
the tributary basins will be important to the overall strategy for managing sedimentation in the Upper
Yangtze. To this end, analysis is first based on tributary groupings: Jinsha±Yalong, Dadu±Min, Jialing
(including Tuo) and Wu. As described above, a large proportion of the catchment areas within the Jinsha±
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1085
SEDIMENT YIELD VARIABILITY
Figure 4. Scattergrams of (a) slope (degrees), (b) annual precipitation (mm) and (c) population density (persons kmÿ2) against elevation
(m) at pixel level
Table III. Spearman's correlation matrix of catchment variables (n = 62)
ME
BR
MS
PD
PP
RO
SY
ME
BR
MS
1
051**
063**
ÿ091**
ÿ046**
ÿ021
ÿ031*
1
042**
ÿ036**
ÿ043**
ÿ014
ÿ005
1
ÿ070**
ÿ017
ÿ033**
021
PD
PP
1
047**
014
023
1
067**
011
RO
SY
1
039**
1
* Significant at 95 per cent level (critical value = 0255)
** Significant at 99 per cent level (critical value = 0335)
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1086
X. LU AND D. L. HIGGITT
Yalong and Dadu±Min tributaries is located in areas with elevation > 3000 m and low population density,
while the Jialing and Wu, particularly their lower branches, are located in areas of lower elevation and high
population density. However, the results of the analysis must be treated with caution because the group of
stations for the Wu tributary contains only seven data points (the six stations listed in Table II plus Shizu (no.
242) which is a small Main Channel tributary close to the Wu±Yangtze confluence).
Specific sediment yield is plotted against all the variables in Figure 5. Generally, wide scatter remains in
the data, despite the tributary groupings. Previous studies have used various strategies to reduce scatter.
Milliman and Syvitski (1992) removed outliers defined by degrees of standard deviation and recalculated
regression equations. Summerfield and Hulton (1994) argued that this procedure masked the real degree of
scatter and proposed logarithmic transformation before calculation of Pearsonian correlation coefficients. In
this study, instead of transforming the variables, curve-fitting techniques have been applied to the data in this
study, with trendlines added to Figure 5 where relationships are significant at the = 005 level. A number of
the relationships between sediment yield and catchment variables are adequately described by second-or
third-order polynomial equations, indicating the apparent presence of turning points or threshold conditions
in the relationship.
There are marked contrasts in the range of significant correlations within each tributary grouping and in the
nature of those relationships. The Jinsha±Yalong tributary has significant correlation with all variables except
for basin relief. The decline of sediment yield with elevation may, at first, appear surprising but represents the
transition from the high altitude, semi-arid Qinghai±Tibet Plateau to the deeply incised, sub-humid valleys
downstream. Sediment yields increase consistently with runoff and precipitation. The Dadu±Min and the Wu
groups have fewer significant correlations ± precipitation and runoff in the former and only runoff in the
latter. The Jialing has the same range of significant correlations as the Jinsha±Yalong, but there is some
difference in the nature of the relationships. Sediment yields increase with elevation up to about 2000 m.
Perhaps surprisingly, population density has an inverse relationship with sediment yield, although a greater
impact through soil erosion might have been expected. Possible explanations are that the highest levels of
population density within the catchment occur within the flat Chengdu Plain and that the more densely
populated areas tend to have higher concentrations of water conservancy structures, which act to trap and
store sediment. The polynomial relationships with precipitation and runoff are interesting. The apparent peak
in sediment yield at an annual precipitation of 500 mm (or runoff around 250 mm) is similar to the classic
model of Langbein and Schumm (1958), while the increase beyond 1000 mm precipitation (or 700 mm
runoff) has some resonance with the global relationship described by Wilson (1973) and Walling and Webb
(1983). However, this pattern is not repeated in the other tributaries. It should also be noted that some
particularly high levels of sediment yield in the upper Jialing are associated with neotectonics and loess soil.
For example, Liuanyang catchment (no. 140) recorded a 30-year average annual specific sediment yield of
about 1700 t kmÿ2 aÿ1 with a maximum exceeding 6600 t kmÿ2 aÿ1 (Table II).
Basin grouping based on size
Analysis of the data grouped by tributary has some utility for recognizing the variation in controlling
factors between different geographical locations and providing guidance for the implementation of
conservancy operations, but relationships are obscured by scale effects reflected in relationships between
specific sediment yield and basin size. As a compromise, the sub-catchments were grouped into four
categories based on their size: <1000 (six catchments), 1000±10 000 (20 catchments), 10 000±100 000 (24
catchments) and >100 000 km2 (12 catchments). It was found that the relationships between sediment yield
and catchment variables improved as catchment size increases (Figure 6). There are no significant
correlations in the <10 000 km2 category. Better correlations are expected for larger catchments as variations
due to land use, soil or geology are minimized. The >100 000 km2 group has strong, positive relationships
with relief, population density and precipitation with a decline in sediment yields with elevation. Here again
there are curious contrasts in the relationships with precipitation and runoff within the different groupings. A
distinct minimum turning point at around 500 mm runoff is depicted by the polynomial trendline for the 10
000±100 000 km2 catchments. There is also a general decline in sediment yield with precipitation within this
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
Copyright # 1999 John Wiley & Sons, Ltd.
SEDIMENT YIELD VARIABILITY
1087
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
Figure 5. Relationships between specific sediment yield and catchment variables for the data grouped by tributary
1088
Copyright # 1999 John Wiley & Sons, Ltd.
X. LU AND D. L. HIGGITT
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
Figure 6. Relationships between specific sediment yield and catchment variables for the data grouped by catchment size
SEDIMENT YIELD VARIABILITY
1089
group in contrast to the relationship for larger catchments. Given the small sample sizes involved, the
influence of outliers on the relationships may be significant.
Basin grouping based on maximum elevation
The importance of elevation as a control on other catchment characteristics was noted earlier. Milliman and
Syvitski (1992) used maximum elevation groupings to demonstrate the relative importance of small
mountainous basins in supplying a large proportion of terrestrial sediment to the oceans. The four classes used
here are: <2500 (eight catchments), 2500±3500 (13 catchments), 3500±5000 (18 catchments) and >5000 m
(23 catchments). The large number of catchments with maximum elevation greater than 5000 m contrasts
starkly with those used in previous studies (Milliman and Syvitski, 1992; Probst and Amiotte-Suchet, 1992;
Summerfield and Hulton, 1994), where the distribution is heavily skewed towards lower elevations.
Relationships between sediment yield and catchment variables are highly scattered within lower elevation
groupings (Figure 7). Sediment yields in lower elevation areas are likely to be influenced by agricultural
activity that is not accounted for in the selected catchment variables. Stronger relationships are found in
the >5000 m grouping, with the familiar inverse relationship evident between sediment yield and mean
elevation. Positive relationships are found with population density, precipitation and runoff.
DISCUSSION: DISCRIMINATING CONTROLS ON REGIONAL SEDIMENT YIELD
Previous attempts to explain sediment yields in terms of a number of controlling factors have encountered
high degrees of scatter. In this study, grouping of catchments by tributary, size and maximum elevation has
provided a means of reducing scatter. Grouping has allowed the controlling factors on sediment yield to be
identified and has revealed that the nature of the resulting relationships can be quite varied. Explanations for
diversity in the relationships between sediment yield and catchment variables within the Upper Yangtze are
summarized below.
Grouping by tributary provided a number of significant relationships for two tributary groups. Multiple
regression using six catchment variables indicates that 87 and 95 per cent of the variance in sediment yields
can be explained in the Jinsha±Yalong and Dadu±Min, respectively. In contrast, only 33 per cent of the
variance is explained in the Jialing. This contrast reflects the larger human impact in the Jialing. The six
selected catchment variables reflect topographic and climatic variables which can be extracted from global
data sets. Population density is the only variable that relates to human impact. Clearly, the inclusion of land
use, soils and geological information would greatly strengthen attempts to model regional sediment yields.
Unfortunately, the resolution of digital information on these variables available within the public domain is
not yet sufficient to support extraction of necessary catchment variables. As land use, geology and soil
impacts are relatively local, analysis of larger catchments generates clearer results than analysis of smaller
ones. The strong, inverse relationship between sediment yield and mean elevation in the >100 000 km2
category reflects the fact that elevation is acting as a surrogate for low precipitation, runoff and human
impact. The nature of the relationship between elevation and sediment yield contradicts all previous studies of
regional or global sediment yields (e.g. Milliman and Syvitski, 1992) but can be attributed to the particular
geography of the Upper Yangtze. The high elevation portions of the catchment are areas of gentle relative
relief, limited precipitation and low human activity. Consequently, fluvial erosion in most of Tibet is limited
and does not export much sediment beyond the plateau (Fielding et al., 1994). Comparatively severe soil
erosion on agricultural land, particularly around the margin of the Sichuan Basin, means that sediment yields
are relatively high at elevations below 5000 m.
Population density can be used as an indicator of the extent and intensity of human impact within the Upper
Yangtze Basin. Its relationship to sediment yield is positive in catchments with higher maximum elevations
and larger areas, but is otherwise scattered. In the Jialing tributary there is a negative relationship between
sediment yield and population density. The variability of this relationship deserves further consideration in
four main respects. First, it should be noted that the population density data are derived from the 1992 census,
while the sediment yield data are averages from the period 1956±1987. During the last 40 years there have
been significant changes in the number and distribution of people living in the Upper Yangtze Basin, so that
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1090
Copyright # 1999 John Wiley & Sons, Ltd.
X. LU AND D. L. HIGGITT
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
Figure 7. Relationships between specific sediment yield and catchment variables for the data grouped by maximum elevation
SEDIMENT YIELD VARIABILITY
1091
the statistical relationships must be treated with caution owing to non-stationarity in the population data.
Second, population density is in any case only a crude measure of human impact. The largest concentrations
of population are in the relatively flat areas of the Sichuan Basin, whereas the most severe soil erosion occurs
on the sloping land around the margins of this area. Third, much of the reported increase in soil erosion in
central China during the measurement period is associated with deforestation and/or expansion of agricultural
activities. By its nature, this tends to be in areas adjacent to high population densities. Further work is
required to model the temporal impact of land disturbance on sediment yield. Fourth, the measurement period
has coincided with a phase of widespread construction of water conservancy structures. Much of the sediment
derived from soil erosion in the higher population areas of the Upper Yangtze may be being trapped in
temporary storage in ditches, ponds and reservoirs. Indeed, storage in small reservoirs may account for the
discrepancy between estimates of increasing soil erosion in the Upper Yangtze and the lack of an upward
trend in sediment yield measured at Yichang (Higgitt and Lu, 1996).
Generally, sediment yield increases with precipitation and runoff in most groupings but displays
polynomial relationships in the Jialing tributary and in catchments in the 10 000±100 000 km2 size category.
This variation reaffirms the contention that no simple relationship exists between these variables (Walling
and Webb, 1983). A general increase with precipitation is consistent with other regional and global analyses
(Probst and Amiotte-Suchet, 1992; Ludwig and Probst, 1996). Xu (1994), using data from 700 rivers in
China, suggested sediment yield attains a maximum for around 400 mm of runoff. This implies that sediment
yields in China fit the Langbein and Schumm (1958) model and are therefore primarily a natural
phenomenon. There is little support for this conclusion in the data for the Upper Yangtze. Multiple regression
results suggest that much of the sediment yield variability in the west of the basin can be attributed to natural
phenomena, but that topographic and climatic variables afford relatively little explanation of sediment yields
in the agricultural, eastern portion of the basin.
The increasing availability and improving resolution of global, environmental databases offers the prospect
of examining the relationships between fluvial sediment yield and various catchment properties within a GIS
framework. To date, GIS-based studies have tended to focus on either global-scale variations or comparisons
between different catchments in a regional context. The analysis of variation within large catchments offers
considerable potential for the investigation and management of sediment-related problems. In the case of the
Upper Yangtze Basin, the attention of environmentalists worldwide has been drawn to the region by the Three
Gorges Project. Integration of sediment yield records and environmental databases within a GIS not only
provides a basis for the empirical investigation of controls on the patterns of sediment yield but also provides
a platform for predictive modelling. However, procedures for dealing with hierarchical data from a series of
nested catchments must be enhanced to deal with a number of problematic issues. Perhaps the most pressing
issues concern development of improved approaches to standardizing specific sediment yields to account for
catchment size and scale effects. Also, mapping conventions for representing sediment yield within a series of
nested catchments are problematic and include shading polygons of incremental catchment area, point
interpolation procedures, predictions from regression models or residuals from whole catchment yield±area
relationships (Higgitt and Lu, 1996; Lu and Higgitt, 1998b)
CONCLUSION
The utility of GIS to extract spatially distributed data for the analysis of intra-catchment sediment yields has
been demonstrated. At the time of writing detailed information on land use, geology and soils is not in the public
domain and the examination of sediment yield variability in the Upper Yangtze Basin has been confined to six
variables: mean elevation, basin relief, mean slope, mean annual runoff, mean annual precipitation and
population density. Several of these variables are strongly correlated with one another, necessitating caution
when interpreting the results of multiple regression analyses. A high degree of scatter in the sediment yield data
results from the diverse characteristics of the catchment, and this has been reduced by grouping the data into
categories based on geographic location, catchment size and maximum elevation, prior to analysis.
The results generally indicate that specific sediment yields increase with precipitation, runoff, population
density and mean slope, but decrease with elevation. Elevation is strongly correlated with other catchment
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
1092
X. LU AND D. L. HIGGITT
variables and is therefore an influential factor on sediment yields, although it has little direct effect on erosion
dynamics. The geography of the Upper Yangtze catchment means that high elevation areas are mostly flat,
semi-arid and sparsely populated, with limited agricultural activity. Within the bivariate plots of sediment
yield against catchment variables some interesting contrasts emerge, particularly with regard to precipitation
and runoff. Of particular significance is apparent scale dependency in the nature of the relationship for
sediment yield as a function of elevation, precipitation and runoff. Further attention to removing scale effects
from the analysis of controls on sediment yields is required.
The limited number of variables used in this study are capable of explaining the majority of the variance of
the measured specific sediment yield for the Jinsha±Yalong and Dadu±Min tributaries, but perform less well
in the more heavily populated areas. The results indicate that topographic, climatic and crude human impact
variables provide a reasonable explanation of sediment yield variations in comparatively natural
environments, but are inadequate in regions influenced by large-scale agricultural activity. It is envisaged
that the resolution of global data sets will continue to improve, allowing the incorporation of variables
representing geological, soil and human activity conditions into regional studies of sediment yield.
ACKNOWLEDGEMENTS
The work was undertaken while the first author was in receipt of an Overseas Research Studentship and
Department of Geography Scholarship at the University of Durham. Xixi Lu would like to thank the British
Geomorphological Research Group and the Great Britain±China Centre for providing financial support to
attend the IAG Bologna Conference. We would like to thank Des Walling and an anonymous referee for some
useful comments and the staff of the Cartography Unit at Durham for tidying up the maps.
REFERENCES
Douglas, I. 1967. `Man, vegetation and the sediment yield of rivers', Nature, 215, 925±928.
Edmonds, R. L. 1992. `The Sanxia (Three Gorges) Project: the environmental argument surrounding China's super dam', Global
Ecology and Biogeography Letters, 2, 105±125.
ESRI 1994. Cell-based modelling with GRID, ArcDoc version 702 on-line help.
Fielding, E., Isacks, B., Barazangi, C. and Duncan, C. 1994. `How flat is Tibet?', Geology, 22, 163±167.
Gu, H. Y. and Douglas, I. 1989. `Spatial and temporal dynamics of land degradation and fluvial erosion in the middle and upper
Yangtze River basin, China', Land Degradation and Rehabilitation, 1, 217±235.
Gu, H. Y., Ai, N. S. and Ma, H. L. 1987. `Sediment sources and trend of sedimentation in the Three Gorges reservoir area', in Leading
Group of the Three Gorges Project Ecology and Environment Research Project, Chinese Academy of Sciences (Ed.), Collected
Papers on Ecological and Environmental Impact of The Three Gorges Project and Countermeasures, Science Press, Beijing, 522±
541 (in Chinese).
Higgitt, D. L. and Lu, X. X. 1996. `Patterns of sediment yield in the Upper Yangtze Basin, China', in Erosion and Sediment Yield:
Global and Regional Perspectives (Proceedings of the Exeter Symposium, July 1996), IAHS Publication 236, 205±214.
Jansen, J. M. L. and Painter, R. B. 1974. `Predicting sediment yield from climate and topography', Journal of Hydrology, 21, 371±380.
Jansson, M. B. 1988. `A global survey of sediment yield', Geografiska Annaler, 70A, 81±98.
Lajczak, A. and Jansson, M. B. 1993. `Suspended sediment yield in the Baltic drainage basin', Nordic Hydrology, 24, 31±52.
Langbein, W. B. and Schumm, S. A. 1958. `Yield of sediment in relation to mean annual precipitation', Transactions of the American
Geophysical Union, 39, 1076±1084.
Legates, D. R. and Willmost, C. J. 1992. `Monthly average surface air temperature and precipitation. Digital Raster Data on a 30
minutes Geographic (lat/long) 360*720 grid', in Global Ecosystems Database Version 10, Disc A, NOAA National Geophysical
Data Centre Boulder, CO.
Lu, X. X. and Higgitt, D. L. 1998a. `Recent changes of sediment yeild in the Upper Yangtze, China', Environmental Management, 22,
697±709.
Lu, X. X. and Higgitt, D. L. 1998b. `Sediment yield mapping using GIS in the Upper Yangtze', in Cao, Z. H. (Ed.), International
Symposium on Soil, Human and Environment Interactions (Nanjing, China, 1997), China Science Technology Press, Beijing, 40±
50.
Ludwig, W. and Probst, J. L. 1996. `A global modelling of the climatic, morphological, and lithological control of river sediment
discharges to the oceans', in Erosion and Sediment Yield: Global and Regional Perspectives (Proceedings of the Exeter Symposium,
July 1996), IAHS Publication 236, 21±28,
Milliman, J. D. and Syvitski, J. P. M. 1992. `Geomorphic/tectonic control of sediment discharge to the ocean: the importance of small
mountainous rivers', Journal of Geology, 100, 525±544.
NOAA-EPA Global Ecosystem Project 1992. Global Ecosystems Database Version 10, User's Guide, Documention, Reprints, and
Digital Data on CD-ROM, USDOC/NOAA National Geophysical Data Center, Boulder, CO.
Ozturk, F. 1996. `Suspended sediment yields of rivers in Turkey', in Erosion and Sediment Yield: Global and Regional Perspectives
(Proceedings of the Exeter Symposium, July 1996), IAHS Publication 236, 65±71.
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)
SEDIMENT YIELD VARIABILITY
1093
Probst, J. L. and Amiotte-Suchet, P. 1992. `Fluvial suspended sediment transport and mechanical erosion in the Maghreb (North
Africa)', Hydrological Sciences Journal, 37, 621±637.
Qian, N., Zhang, R. and Chen, Z. C. 1993. `Some aspects of sedimentation at the Three Gorges Project', in Luk, S-H. and Whitney, J.
B. (Eds), Megaproject A, Case Study of China's Three Gorges Project, M.E. Sharpe, Armonk, NY, 121±160.
Ruddiman, W. F., Prell, W. L. and Raymo, M. E. 1989. `Late Cenozoic uplift in southern Asia and the American West: Rationale for
general circulation modeling experiments', Journal of Geophysical Research, 94, 18, 379±18, 391.
Summerfield, M. A. and Hulton, N. J. 1994. `Natural controls of fluvial denudation rates in major world drainage basins', Journal of
Geophysical Research, 99(B7), 13, 871±13, 883.
Walling, D. E. 1983. `The sediment delivery problem', Journal of Hydrology, 65, 209±237.
Walling D. E. 1983 `Patterns of sediment yield', in Gregory, K. J. (ED), Background to palaeohydrology, Wiley, Chichester 65±100.
Waythomas, C. F. and Williams, G. P. 1988. `Sediment yield and spurious correlation ± Toward a better portrayal of the annual
suspended sediment load of rivers', Geomorphology, 1, 309±316.
Wen, D. Z. 1993. `Soil erosion and conservation in China', in Pimental, D. (Ed.), World Soil Erosion and Conservation, Cambridge
University Press, Cambridge, 63±85.
Wilson, L. 1973. `Variation in mean annual sediment yield as a function of mean annual precipitation', American Journal of Science,
273, 335±349.
Xu, J. X. 1994 `Zonal distribution of river basin erosion and sediment yield in China', China Science Bulletin, 39, 1356±1361.
Zhou, G. Y. and Xiang, Z. 1994. `Sediment transport in the Yangtze basin', in Variability in Stream Erosion and Sediment Transport
(Proceedings of the Canberra Symposium, December 1994), IAHS Publication 224.
Copyright # 1999 John Wiley & Sons, Ltd.
Earth Surf. Process. Landforms 24, 1077±1093 (1999)